Nevertheless, we must be cautious about these comparisons because part of our experimental sample is initially covered by a CHI plan. We discuss this point below.
In addition to the weakness of the takeup rate elasticity, the low takeup rate is intrigu- ing. We explore other potential explanations and ask whether the imprecise eligibility
assessment rule used by the CPAM partly explains this weak enrollment rate.
A. Takeup rate and eligibility
As reported in Table 7, a sizeable number of applications were refused by the CPAM due to resources that were too high or too low. This imprecise targeting of the eligible
population may be an initial explanation for low enrollment because noneligible in- dividuals may know that they cannot benefi t from ACS for instance, through former
social benefi ts applications.
Our experiment underlines the diffi culties in reaching the targeted population. These
Table 8 Likelihood of ACS Takeup
Dependent Variables Average Marginal Effect
Group Control
Reference Treatment
1 0.025
0.0139 Treatment
2 –0.004
0.0137 Age
0.012 0.0021
Age
2
–0.0001 0.0002
Female –0.026
0.0115 Employment status in 2008
Working individual
Reference Disabled
individual 0.154
0.0342 Retired
individual 0.152
0.0187 CMU- C coverage in 2007
0.028 0.0236
CHI coverage in 2008 0.017
0.1187 Long- term illness in 2008
–0.007 0.0161
Ambulatory healthcare expenditures in 2008 0€–200€
–0.054 0.0177
200€–699€ –0.007
0.0184 700€–1,999€
0.011 0.0167
2,000€ or more Reference
Pseudo R
2
0.0613 N
4,209
Notes: Probit regression of the probability of ACS takeup dummy variable: 1 individual returned an ap- plication form; 0 otherwise. Average marginal effects are reported and standard errors are in parentheses.
Statistical signifi cance levels =10 percent; =5 percent; =1 percent.
diffi culties are due to the administration’s lack of precise information on actual family resources; thus, we do not have accurate information on who is eligible for ACS. As
mentioned above, a nonnegligible number of individuals who returned an application form were not eligible for ACS. It is essential to take this factor into account because
uncertainty about effective eligibility reduces the incentive to apply for the program.
To further investigate this issue, after the experiment, we collected new data on 2008 incomes for each individual to more precisely assess eligibility for ACS in 2009.
After reassessing eligibility with this new information, we found that only 43 percent of the experimental sample for whom 2008 income was available, were eligible for
ACS, confi rming that the initial eligibility targeting of the experimental population in 2008 was imprecise Table 9. All the fi gures presented in this section are dragged
from this new defi ned sample.
Among this redefi ned eligible population, the takeup rate rises to 24 percent. The failure of some individuals in the initial sample to apply could be explained
by their knowledge of their ineligibility status. However, this rate remains low and suggests other obstacles to application. The refusal rate among applicants remains
high: 22.3 percent of the claimants were refused due to resources beyond the entitle- ment criteria, of which 5 percent were refused due to resources that were too low
and 17.2 percent because of resources that were too high. This fi nding underlines the complexity of eligibility criteria, which is heightened by the narrowness of the tar-
get population. Moreover, imprecise targeting of eligible people implies costs for the CPAM: the direct costs of sending letters to noneligible people, indirect costs due to
the nonapplication of eligible individuals because of uncertainty about their eligibility, and the nonfi nancial costs borne by noneligible people who apply.
Our results also suggest that health insurance subsidies could help to better target eligible populations given that the acceptance rate is slightly higher in both treatment
groups than in the control group 79.2 percent versus 74.4 percent. In particular, the proportion of refusals due to resources above the eligibility threshold was lower in
both treatment groups, suggesting that the increase in the subsidy has attracted the poorest among the experimental sample.
B. Information costs